The present invention relates to a template matching technique, and more particularly to the template matching technique which is effective especially when a region corresponding with an object of an input image is more transformed as compared with an object region of a template image or when noise is superposed on the input image.
There has been conventionally known the technique called the template matching. This template matching technique is arranged to take the steps of creating a template image from an object image taken by an imaging device, obtaining an evaluated value of matching between an input image and the template image (which value is called a degree of matching or matching degree), and determining if the input image corresponds to the object image or if the object image is included in the input image based on the obtained matching degree.
For example, the basic method of the template matching is described in the publication of “Kouichi SAKAI “Fundamentals and Applications of Digital Image Treatment by Visual C#.NET & Visual Basic.NET—From Basic Concept to Face Image Recognition”, CQ Publishing House, First Edition issued on 1 Sep. 2003, Chapter 5, Section 3, pp. 84-86”. In FIGS. 5 and 6 shown in this publication, an image of a girl's face is used as an original image. The technique is arranged to take the steps of cutting a portion of a left eye from the image as a template, creating an input image by executing the density conversion of the original image, executing the correlating operations of portions of the input image with respect to the template image, and showing the result. In this example, when a correlation coefficient has a threshold value of 0.99, only the left eye portion cut as the template image corresponds to the matched position, while when the coefficient has a threshold value of 0.8, several matching candidate positions are detected in not only the left eye portion but the portions around the right eye.
As to the template matching technique, there have been several methods for lessening a time of a matching operation. For example, JP-A-63-211474 discloses the technique of hierarchically reducing a size of a template image, a sampling rate and a candidate region for the purpose of enhancing the operating speed more than the template matching to be executed with the template image not to be reduced hierarchically. Further, for example, as to the digital image, the foregoing publication discloses the technique of speeding up the operation time by using a factor of “distance” as an evaluating expression if a subtraction and a normalization based on an average value may be left off. Moreover, for example, if an object to be recognized has a relatively simple form, the Official Gazette of JP-A-2001-351102 discloses the technique of speeding up the matching operation of a digital image by using as a template a circle whose diameter is a width of a circumscribed quadrangle composing the object to be recognized.
The template matching technique is now applied into the wide range of appliances such as various checking devices, a recognizing device, a determining device and a positioning device.
For example, JP-A-8-181053 discloses the positioning method for an exposing device which method is arranged to use the template matching. If a portion of a template image with lots of errors are grasped in advance, the method is executed to take the steps of masking the portion and the region of the imaged image (input image) corresponding with the masked portion of the template image and executing the predetermined evaluating operation with respect to the unmasked region only. This method makes it to possible to exclude the portion with a large error when evaluating the input image, thereby leading to enhancing the positioning accuracy.
Moreover, JP-A-2007-5818 discloses the template matching technique arranged to use the data designed by a CAD (Computer-Aided Design) system when designing a semiconductor. Concretely, the image obtained by imaging the manufactured semiconductor with an electron microscope is used as the input image. Then, the template image is created by taking the steps of creating a bitmap image based on semiconductor design data and smoothing the bitmap image. This template image and the input image are used for the template matching process.